>> Is this block layout efficient in every matrix operation? >> How to choose the block size to make the parallel computation efficient?
Good point. We should think/discuss about it. However, We have a capacity of 2d blocking at this time. :) On Tue, Oct 21, 2008 at 11:03 AM, Samuel Guo <[EMAIL PROTECTED]> wrote: > hmm, your explanation is clear. > > but if we store the matrix to file, the block layout is fixed. > Is this block layout efficient in every matrix operation? > How to choose the block size to make the parallel computation efficient? > > On Tue, Oct 21, 2008 at 9:36 AM, Edward J. Yoon <[EMAIL PROTECTED]>wrote: > >> For example, >> We have a 4 * 4 Matrix a : >> >> [a, b, c, d] >> [e, f, g, h] >> [i, j, k, l] >> [m, n, o, p]] >> >> Then, we can store an 2 * 2 block matrices to file. >> >> block(0, 0) is a 2 * 2 sub-matrix as below. >> >> [a, b] >> [e, f] >> >> Key : block(0, 0), Value : submatrix [a, b], [e, f] >> Key : block(0, 1), Value : submatrix [c, d], [g, h] >> Key : block(1, 0), Value : submatrix [i, j], [m, n] >> Key : block(1, 1), Value : submatrix [k, l], [o, p] >> >> ..... >> >> Also, If we can access sub-matrix from Hbase table, blocking can be >> pre-computed. >> >> On Tue, Oct 21, 2008 at 10:18 AM, Samuel Guo <[EMAIL PROTECTED]> wrote: >> > should the matrices that do the multiplication have the same blockID >> layout? >> > >> > And How to divide a matrix into BlockIDs to make the parallel computation >> > most efficiently? >> > >> > 2008/10/21 Edward J. Yoon <[EMAIL PROTECTED]> >> > >> >> Can anyone review this ?? >> >> >> >> http://blog.udanax.org/2008/10/parallel-matrix-multiply-on-hadoop.html >> >> >> >> -- >> >> Best regards, Edward J. Yoon >> >> [EMAIL PROTECTED] >> >> http://blog.udanax.org >> >> >> > >> >> >> >> -- >> Best regards, Edward J. Yoon >> [EMAIL PROTECTED] >> http://blog.udanax.org >> > -- Best regards, Edward J. Yoon [EMAIL PROTECTED] http://blog.udanax.org
